Self Code Review
Makes the agent review its own code before you ship it. Six passes, findings ranked by real impact, and a verdict it is allowed to answer with 'this is clean'.
- Claude Code
- Open Source
- Skill
an agent grading its own homework fails in one of two directions. it rubber stamps, because the code it just wrote obviously made sense to it. or it manufactures a wall of nitpicks, because a review that found nothing feels like a review that did not happen.
both are useless. this forbids both.
it defines what a real finding actually is. "this could be better" is not a finding, it is a feeling. "lines 42 to 58 duplicate the validation logic from lines 12 to 28 in auth.ts" is a finding. name the file, the line, what is wrong, why it matters.
it ranks by impact, not by what is easiest to spot. a review that opens with a style nit while a bug sits on line 42 has failed.
and clean is a first class verdict. if the code is good it says so and stops, instead of inventing work to justify the invocation.
the part i like most is the "no action needed" section at the end. it is the difference between "i found nothing" and "i checked these seven things and they are fine".